Design and Analysis with Asynchronous Information


Agents within a network structure can benefit from shared information, but will also likely need to make independent decisions (i.e., traditional centralized control is unlikely to succeed in an adversarial environment). Based on locally sensed or communicated information, agents may adjust the mission plan (e.g., identifying a nearby adversary and switching from a patrol mode to a pursuit mode). That is, each agent in a network will make onboard decisions to switch between different sensing or control actions (i.e., discrete events in the hybrid dynamics), and the timing and trigger for each switch will be informed by data an agent gathers itself and data it receives from other agents in the network, all of them likely at different time instances. This naturally leads to hybrid models and demands new tools for hybrid systems. Yet, delays are inevitable in adversarial environments, and the effect of such delays impact the function of an agent. Efforts focus on generalizing the analysis tools for the design of networked hybrid systems with heterogeneous delays of unpredictable length on the triggering events, such as those associated with communication of information, reconfiguration of parameters, and change of mode of operation. Analyzing the stability of such systems will require both novel analyses of hybrid systems themselves and fundamentally new approaches to understanding how online data processing and computation affects controllers. Another major consequence of online computation, including cyber attack-detection and resiliency measures is that control decisions will rely on the intermediate iterates of their computations, rather than merely the final outputs of these computations. Asynchronous computations then not only impact each agent's computations, but also their control decisions. These effects can be seen in the algorithms agents execute, and we develop convex optimization methods that consider the breadth of computational and data processing tasks they encapsulate. Specifically, first-order primal-dual methods are developed both for the computational simplicity they provide and their ability to accommodate multi-agent implementations. We design such methods to provide robustness to the asynchronous and noisy data sharing encountered in adversarial settings.


Design and Analysis with Asynchronous Information Publications